The Independent

AI can easily be trained to lie – and it can’t be fixed, study says

Source: Reuters

Advanced artificial intelligence models can be trained to deceive humans and other AI, a new study has found.

Researchers at AI startup Anthropic tested whether , such as its Claude system or OpenAI’s ChatGPT, could learn to lie in order to trick

You’re reading a preview, subscribe to read more.

More from The Independent

The Independent4 min read
Man City Can Eclipse Manchester United Thanks To Sir Alex Ferguson’s Famous Habit
The time has passed when Manchester United were the permanent reference point for Manchester City. If they used to be role models or targets, now they can seem more of a salutary warning. But at a point when City are four games from displacing United
The Independent3 min read
Heathrow UK Border Force Strike Begins – But No One Seems To Notice
The first arrivals at London Heathrow on day one of a strike by UK Border Force have reported no problems getting through passport control. More than 300 PCS union members working for UK Border Force have walked out for four days, from 29 April to 2
The Independent1 min read
Pair Of Giant Pandas Set To Travel From China To San Diego Zoo Under Conservation Partnership
A pair of giant pandas will soon make the journey from China to the U.S., where they will be cared for at the San Diego Zoo as part of an ongoing conservation partnership between the two nations, officials said Monday. The San Diego Zoo Wildlife All

Related Books & Audiobooks